Course topics
Part 1. Correspondence of Local Features
- Local features: introduction, terminology
- Motivation: a generalization of local stereo to wide-baseline stereo
- Detection of Local invariant features
- Descriptors, Matching
- Correspondence Verification
- RANSAC (robust model fitting)
Part 2. Tracking
- Template Matching and the Lucas-Kanade Method
- Mean Shift HW
Part 3. Graphical Models
- Probabilistic models. Decisions under uncertainty
- Hidden Markov Model. Markov chain, different factorizations. MAP problem — Viterbi algorithm
- Marginals problem — forward-backward algorithm
- Generalized dynamic programming
- Markov Random Field. MAP in MRF
- Energy minimization, solvable classes, graph cuts, relaxations
- Graphical models as neural networks. Sigmoid Belief network
- Uncertainties, noises
- Variance propagation methods
- Bayesian Learning. Variational Bayesian learning
Part 4. Geometry
- Perspective camera model and calibration
- Homography between two images
- Projection and homography
- Epipolar geometry and camera motion
- 3D Reconstruction
- Epipolar geometry and 3D reconstruction
Part 5. Retrieval
- Image Retrieval
- K-Means
- Min-hash
Part 6. Deep Learning for Computer Vision
Part 7. Advanced CNN Topologies